Project Description
Machine learning (ML) based algorithms which identify heavy objects (e.g. b quarks and Z bosons) in jets, e.g. deep neural networks (DNNs), have been established firmly offline but not yet used in real time analyses (RTA). As a first objective, the student in this project will understand the resource cost of these DNNs, improve them if necessary, and adapt them to be used for real time analyses in the general purpose RTA stream for Run 3. This will significantly extend the applicability of the RTA stream for analyses. The industrial secondment to VERIZON will have the second objective of adapting ML frameworks, e.g. TensorFlow Lite, in resource-constrained environments, for real-time processing of images captured by embedded devices. This is an instrumental step towards in-vehicle embedded computing applications for VERIZON, where the student will be trained in theory and best practices of ML, and will return with expertise in RTA ML frameworks in constrained environments to improve the initial trigger selections towards their application in physics analysis. The third objective of this project is to exploit ML in trigger algorithms to validate the generic RTA stream by measuring the frequent and well understood Z → b b process, and then apply it to a new measurement of the H → b b process. The student will also collaborate with CERN experts and work on the monitoring, improvement, exploration, and analysis of the generic RTA stream. The fourth objective towards the end of Run 3 data-taking is to establish the general purpose RTA stream for High-Luminosity LHC, exploiting the extended capabilities of the new detector (tracking at Level-1 trigger, high granularity calorimeter, timing information).
Host country: Finland
Host beneficiary: University of Helsinki
PhD-awarding institution: University of Helsinki
Planned collaborations: VERIZON, CERN